WEBVTT - UL NO. 485: STANDARD EDITION: Netflix RCE, My Current AI Stack, All-in on Claude Code, and more... 0:00:17.267 --> 0:00:20.267 All right. Welcome to episode 45. Posted a bunch of 0:00:20.267 --> 0:00:23.347 fabric extractions of stuff that I've been watching and reading 0:00:23.347 --> 0:00:27.227 and stuff like that. So like YouTube videos, blog posts, 0:00:27.627 --> 0:00:30.107 academic papers, and I think I'm going to do more 0:00:30.107 --> 0:00:33.107 of this because I'm actually building this into my Pi, 0:00:33.147 --> 0:00:37.107 which is pi personal AI infrastructure. And this is the 0:00:37.107 --> 0:00:40.307 infrastructure you've heard me talking about in the augmented class. 0:00:40.987 --> 0:00:43.987 And just like all over X and in the newsletter 0:00:43.987 --> 0:00:46.987 over the years, this is basically the back end core 0:00:46.987 --> 0:00:51.347 infrastructure that I basically runs for my life. So it's 0:00:51.346 --> 0:00:56.987 basically finding all my content. It's categorizing all my content, labeling, sorting, 0:00:57.027 --> 0:01:02.867 doing things like that, extracting summaries, um, you know, characterizing. 0:01:02.867 --> 0:01:06.147 And I'm basically going to use that to continue enhancing 0:01:06.147 --> 0:01:08.467 my life. Right. So I'm always adding to this thing. 0:01:09.267 --> 0:01:11.947 So one of the core components of that is actually 0:01:13.507 --> 0:01:17.607 finding new content, um, in an automated way. So basically 0:01:17.607 --> 0:01:21.167 there's I'll give you a good example. There's on arXiv. 0:01:21.207 --> 0:01:22.967 I think that's the way you pronounce this. I've heard 0:01:22.967 --> 0:01:27.207 people say this, I used to call it arXiv because whatever. 0:01:27.207 --> 0:01:32.767 I'm silly. arXiv, I guess it's called arXiv. It's a 0:01:32.767 --> 0:01:36.727 cool name. It's a better name than arXiv, although that's 0:01:36.727 --> 0:01:40.206 how it's spelled. But whatever on arXiv there is a 0:01:40.607 --> 0:01:48.287 RSS feed for. CS I so it's papers coming out 0:01:48.287 --> 0:01:52.647 of arXiv that are both CS related and AI related. 0:01:52.687 --> 0:01:55.447 And obviously this is a sweet spot for me. So 0:01:56.047 --> 0:01:57.967 I want to be able to go and get those papers. 0:01:57.967 --> 0:02:01.247 But I also want to know how high quality they are, like, 0:02:01.247 --> 0:02:04.487 what are the interesting claims that they're making? Like how 0:02:04.487 --> 0:02:08.687 good was their methodology, their testing? Like, was it a 0:02:08.686 --> 0:02:12.007 meta study? Did they do their own independent research for 0:02:12.007 --> 0:02:15.937 the paper, like all those combinations. And there's already a 0:02:15.936 --> 0:02:18.337 fabric pattern for this, which is analyze paper. And I 0:02:18.337 --> 0:02:20.417 think I have a new version of it as well. 0:02:20.737 --> 0:02:23.377 But anyway, I want to do that type of automated 0:02:23.377 --> 0:02:27.337 analysis on every paper I might want to read. Okay. 0:02:27.376 --> 0:02:30.096 This is kind of giving you hints of like what 0:02:30.096 --> 0:02:33.897 my personal AI infrastructure does. And by the way, I'm 0:02:33.897 --> 0:02:35.776 aware of the fact that I'm saying like a little 0:02:35.776 --> 0:02:40.376 bit too much. So I'm trying to tamper that down. 0:02:40.417 --> 0:02:43.537 I like the word. I think it's a cool transition word, 0:02:43.536 --> 0:02:45.417 but I don't want to use it 40 times in 0:02:45.417 --> 0:02:50.697 every sentence. Anyway, this is the infrastructure that I'm actually building, 0:02:51.017 --> 0:02:53.577 and this is one of the modules that's really, really 0:02:53.577 --> 0:02:56.416 important to me because it's not just papers. I want 0:02:56.457 --> 0:02:58.417 to see this for open source intelligence. I want to 0:02:58.417 --> 0:03:00.897 see this for all the different stuff. So these are 0:03:00.897 --> 0:03:04.656 all the independent modules that I'm building. And actually those 0:03:04.656 --> 0:03:07.977 essentially turn into products for me. Right. So my Intel 0:03:07.977 --> 0:03:10.897 product is going to be the output of one of these. 0:03:11.137 --> 0:03:15.397 Threshold is already the output of several of these, right? 0:03:15.517 --> 0:03:21.637 So anyway, that's just a brief little thing on my pie. Pie. 0:03:23.957 --> 0:03:27.677 All that to say that these fabric extractions are essentially 0:03:27.677 --> 0:03:31.637 outputs from these modules where it's found a piece of 0:03:31.637 --> 0:03:34.436 really unique content, or I manually found it and I 0:03:34.436 --> 0:03:39.796 just manually ran, uh, fabric against it. And it outputted 0:03:39.797 --> 0:03:42.077 this output so you could check it on X, you 0:03:42.077 --> 0:03:44.197 could see it on, you know, it. I do a 0:03:44.197 --> 0:03:46.517 few of these in the newsletter every once in a while, 0:03:46.517 --> 0:03:50.397 but I find it really useful to go and extract predictions, 0:03:50.637 --> 0:03:54.517 to extract wisdom, to extract insights, or is another one 0:03:54.517 --> 0:03:58.197 of my favorite. There's also one that just extracts quotes, 0:03:58.557 --> 0:04:02.597 which can also be really interesting. There's um, I think 0:04:02.597 --> 0:04:07.197 there's a one that just extracts like data or recommendations. 0:04:07.197 --> 0:04:11.157 I think there's definitely a recommendations. One, actually, now that 0:04:11.157 --> 0:04:13.447 I think about it, I need to have one that 0:04:13.447 --> 0:04:16.847 extracts like a book list or a watch list. Um, 0:04:16.847 --> 0:04:18.607 I'm not sure if I already made that one or 0:04:18.607 --> 0:04:21.487 if I need to go add that to fabric. Anyway, 0:04:21.487 --> 0:04:27.207 all this to say, I want my pi Pi always 0:04:27.207 --> 0:04:31.607 running in the background, always collecting, always finding, always discovering, 0:04:31.607 --> 0:04:36.287 always rating. And before you think, well hold on, I 0:04:36.327 --> 0:04:39.607 is doing too much work for you. No, it is 0:04:39.607 --> 0:04:42.407 not doing too much work for me. And here's why. 0:04:43.687 --> 0:04:48.367 Part of this entire thing is the determination of what 0:04:48.367 --> 0:04:52.207 I should go and do slowly. Okay? Anything that makes 0:04:52.207 --> 0:04:56.487 it into that bucket, I must sit down and manually 0:04:56.487 --> 0:05:00.807 enjoy and listen to and process and use my pen 0:05:00.807 --> 0:05:04.447 and my index cards and write down notes or Apple 0:05:04.447 --> 0:05:06.367 notes or whatever it is if I have to pause 0:05:06.367 --> 0:05:09.607 this thing. But I still think the absolute best way 0:05:09.607 --> 0:05:14.347 to experience anything is in the slow form, right? What's 0:05:14.347 --> 0:05:17.387 still the best possible way to spend an evening is 0:05:17.427 --> 0:05:22.107 with friends and family and smart people or whatever, and 0:05:22.107 --> 0:05:26.827 you're basically having conversation with actual humans. So to me, 0:05:26.827 --> 0:05:29.827 all this AI stuff, the whole purpose of it is 0:05:29.827 --> 0:05:33.387 to enhance your ability to do those things. And this 0:05:33.387 --> 0:05:36.307 is still this is all part of human 3.0. It's 0:05:36.307 --> 0:05:38.987 all part of like, what should we actually be doing? 0:05:39.307 --> 0:05:42.987 And what we should actually be doing is magnifying human experience. 0:05:43.427 --> 0:05:48.627 Slow experience. Right. All this speeding up and all this 0:05:48.627 --> 0:05:52.147 collection and all this, I should be in service of that, 0:05:52.507 --> 0:05:57.827 not in opposition to that. All right. That's how we 0:05:57.827 --> 0:06:00.227 get stuck on the first item. Not even getting into 0:06:00.227 --> 0:06:04.187 the newsletter. Okay. Yeah. Go sign up for the podcast again, 0:06:04.187 --> 0:06:06.787 which is silly because you're already listening to it. Uh, 0:06:06.787 --> 0:06:11.197 let's see here my friend Tracy Talbot is presenting Tell 0:06:11.237 --> 0:06:17.437 No Lies. Teaching AI to know when it doesn't know. Yeah, 0:06:17.437 --> 0:06:19.737 to know when it doesn't know. And this is June 0:06:19.737 --> 0:06:22.877 26th in Austin, and she's got a LinkedIn post that 0:06:22.877 --> 0:06:25.957 I linked in the newsletter. I worked with her at Apple. 0:06:25.957 --> 0:06:31.676 She was extraordinary there at Apple. She was doing data quality, uh, 0:06:31.677 --> 0:06:37.517 on our team. And yeah, just really, really great person and, um, 0:06:37.597 --> 0:06:42.197 really thoughtful about data quality and data security and all 0:06:42.197 --> 0:06:48.197 these different aspects of data and basically data management at 0:06:48.197 --> 0:06:51.557 scale and thinking about what it means to trust data. 0:06:51.997 --> 0:06:56.197 So now she's getting into AI, and she's been doing 0:06:56.197 --> 0:06:57.957 a whole lot of really cool stuff with that. So 0:06:57.957 --> 0:07:01.837 definitely recommend going to see her talk again. That's June 0:07:01.837 --> 0:07:05.797 26th in Austin. All right. Places I'm going to be 0:07:05.797 --> 0:07:09.777 speaking in 2025, and I realized I missed an item 0:07:09.777 --> 0:07:13.657 on the list. Various talks and panels. Blackhat USA in 0:07:13.657 --> 0:07:17.657 Vegas speaking at the Swiss Cyberstorm in Bern, Switzerland. It's 0:07:17.657 --> 0:07:20.777 going to be an AI workshop. Actually, I'm going to 0:07:20.777 --> 0:07:22.937 do a workshop for people, which is going to be 0:07:22.937 --> 0:07:27.697 a combination of like augmented fabric slash, essentially like personal 0:07:27.697 --> 0:07:30.177 AI infrastructure is kind of what that thing's going to 0:07:30.177 --> 0:07:34.817 be about. Keynoting Appsec USA in DC in November and 0:07:34.817 --> 0:07:40.537 speaking at Blackhat Mia in Saudi Arabia in December. And 0:07:40.537 --> 0:07:43.017 the one that I missed is I'm probably going to 0:07:43.017 --> 0:07:47.697 be doing something in Sweden as well. Um, I can't 0:07:47.737 --> 0:07:50.737 remember the exact date for that one, but, uh, yeah, 0:07:51.057 --> 0:07:53.737 I'll probably add that to the list for this next 0:07:53.737 --> 0:07:59.577 coming newsletter. On Monday, personal tech Stack updates I'm all 0:07:59.577 --> 0:08:02.457 in on cloud code. I'm not going to go into why, 0:08:02.497 --> 0:08:08.147 but ultimately it's because here I am going into why, uh, 0:08:08.307 --> 0:08:13.787 anthropic uses this as their agentic coding agent, and they've 0:08:13.827 --> 0:08:18.027 gone all in on Agentic coding, which means this thing 0:08:18.027 --> 0:08:22.427 isn't like hacking together agents. It's actually using agents to 0:08:22.467 --> 0:08:25.907 execute the various tasks. And as of today, I just 0:08:25.907 --> 0:08:29.027 updated cloud code and now it has planning built in. 0:08:29.947 --> 0:08:32.666 So as I've been talking about pretty much everywhere, I 0:08:32.667 --> 0:08:38.667 think scaffolding around AI is the superpower that's about to 0:08:38.707 --> 0:08:42.867 unlock this whole thing, because the models are already smart. 0:08:42.867 --> 0:08:45.547 The problem is they can't think for long. They have 0:08:45.547 --> 0:08:49.666 short memories. So we have all these context problems. So 0:08:50.347 --> 0:08:55.147 any any person, any intelligent being gets really stupid when 0:08:55.146 --> 0:09:00.146 it can't remember longer than, you know, a short conversation. 0:09:00.546 --> 0:09:03.027 And then you have to go back into this stored 0:09:03.026 --> 0:09:06.967 memory bank. And it's not perfect. I'm talking about rag, 0:09:06.967 --> 0:09:10.607 for example. Or if you have actual humans and they 0:09:10.607 --> 0:09:13.327 struggle with long term memory or even short term memory, 0:09:13.766 --> 0:09:16.766 obviously they're going to be less functional, right? They're going 0:09:16.807 --> 0:09:18.927 to be less capable than they used to be before 0:09:18.926 --> 0:09:25.327 they lost those abilities. So the ability to um, as 0:09:25.367 --> 0:09:29.167 Dwarkesh talks about, learn on the job. Okay. You think 0:09:29.166 --> 0:09:32.247 about a knowledge worker. They go in, they do orientation, 0:09:32.247 --> 0:09:34.807 they learn on the job. Okay. Guess what? After that 0:09:34.807 --> 0:09:38.566 first week, they're way better than when they started because 0:09:38.567 --> 0:09:41.806 now they have some knowledge. Well, what happens after six months? 0:09:41.807 --> 0:09:44.567 What happens after a year? What happens after they've been 0:09:44.567 --> 0:09:48.207 there for 20 years. They now have all this stored 0:09:48.207 --> 0:09:52.287 knowledge about this company, which is now part of their context. 0:09:53.247 --> 0:09:57.766 But the difference is with a human, they are naturally, fluidly. 0:09:57.766 --> 0:10:01.887 Every time they make a decision, they are thinking of 0:10:02.016 --> 0:10:06.497 their stored knowledge. They incorporate all of their decisions with 0:10:06.497 --> 0:10:10.417 their stored knowledge every time they make any decision. That 0:10:10.416 --> 0:10:14.416 is a thing that I currently cannot do. It is struggling. 0:10:14.416 --> 0:10:16.576 It's like, oh, I've got to go grab all this 0:10:16.577 --> 0:10:20.296 context and stuff it into the prompt. How do you 0:10:20.296 --> 0:10:24.416 grab context of 25 years of learning at a job? 0:10:25.296 --> 0:10:27.056 How do you know what that even means? How do 0:10:27.057 --> 0:10:32.217 you even know what to query to stuff into your 0:10:32.256 --> 0:10:37.177 decision making process? This is completely transparent with humans. When 0:10:37.176 --> 0:10:40.256 we are like, oh, let me approach this problem from 0:10:40.256 --> 0:10:43.097 a holistic standpoint. Let me talk to talk to all 0:10:43.097 --> 0:10:46.057 these disparate teams across the org. And I'm going to 0:10:46.057 --> 0:10:47.776 gather notes and I'm going to take notes. I'm going 0:10:47.776 --> 0:10:50.577 to write them down or whatever. Your brain is going 0:10:50.617 --> 0:10:53.417 crazy with all of its knowledge that you've gathered over 0:10:53.416 --> 0:10:57.536 these 25 years, and it's making its own queries and 0:10:57.536 --> 0:11:01.597 pulling back that context. And these are coming back to 0:11:01.597 --> 0:11:04.837 you as kind of like insights or thoughts or creativity. 0:11:05.557 --> 0:11:08.156 You're not getting to see exactly what the queries look 0:11:08.156 --> 0:11:12.197 like or even what the returned results look like. We 0:11:12.197 --> 0:11:15.756 do not have this with AI. We must manually go 0:11:16.156 --> 0:11:20.797 and make manual queries into context, like with rag or 0:11:20.837 --> 0:11:25.756 like with stuffing other context or context summaries into the 0:11:25.756 --> 0:11:29.756 prompts themselves, right? This is the scaffolding that we need. 0:11:29.796 --> 0:11:35.277 We need to emulate the process that humans execute naturally 0:11:35.437 --> 0:11:39.237 every day, you know, hundreds or thousands or millions of times. 0:11:39.237 --> 0:11:42.556 However many times it is that we're making decisions and 0:11:42.557 --> 0:11:47.357 little tiny, you know, nano decisions and micro decisions. So 0:11:48.756 --> 0:11:53.357 I don't remember how I got onto this, but scaffolding massively, 0:11:53.357 --> 0:11:57.997 massively important. And, uh, that's why I think context is, 0:11:58.036 --> 0:12:01.646 is so important to all of this stuff. Um. Oh, 0:12:01.687 --> 0:12:05.406 I remember how Claude Cote. So they are all in 0:12:05.406 --> 0:12:08.446 on the agent stuff, and they are all in on 0:12:08.487 --> 0:12:12.047 the planning component. Okay, this planning component is yet another 0:12:12.046 --> 0:12:15.807 piece of what I just talked about. It is, you know, 0:12:15.847 --> 0:12:19.047 keeping in mind what we're trying to do overall and 0:12:19.046 --> 0:12:21.926 then checking off little boxes of like, yeah, we've done this. Yeah, 0:12:21.926 --> 0:12:27.886 we've done this. It's all about, uh, context management. Now 0:12:27.926 --> 0:12:31.326 context is a little bit, um, overloaded here. I would 0:12:31.327 --> 0:12:36.967 say context planning management. Um, learning on the job management, 0:12:37.607 --> 0:12:41.127 long term task management. These are the types of things 0:12:41.127 --> 0:12:47.766 that really good knowledge workers, like a principle engineer who's 0:12:47.766 --> 0:12:51.327 been somewhere for 25 years. These are the types of 0:12:51.327 --> 0:12:54.006 things that they are really good at. And these are 0:12:54.006 --> 0:12:56.847 the things in my opinion, this is a prediction and 0:12:56.847 --> 0:13:01.427 this is just my prediction, is that we're going to 0:13:01.426 --> 0:13:05.307 see major advances in those areas, which are actually kind 0:13:05.307 --> 0:13:08.227 of like tricks. This is what I've been telling my friend, uh, 0:13:08.426 --> 0:13:12.266 Joel ever since 23, that these types of tricks are 0:13:12.266 --> 0:13:14.426 the things that are going to push us over the edge. 0:13:14.627 --> 0:13:16.667 These are the types of tricks that are going to 0:13:17.307 --> 0:13:23.306 hyper jump. I way further ahead. And I think what 0:13:23.347 --> 0:13:26.867 I think my theory has been borne out through things 0:13:26.867 --> 0:13:33.347 like Post-training post-training is not fundamentally better neural nets. It's 0:13:33.347 --> 0:13:38.427 better ways of shaping and getting neural nets to do 0:13:38.426 --> 0:13:43.146 what we actually want them to do. Right. So just 0:13:43.146 --> 0:13:46.627 zooming out, I think Cloud Code has this figured out. 0:13:46.627 --> 0:13:51.067 I think anthropic really deeply understands this. And they are 0:13:51.067 --> 0:13:56.847 building agents and planning management and context Management. They're doing really, 0:13:56.847 --> 0:14:00.207 really well with it in a way that I'm just 0:14:00.247 --> 0:14:02.967 really excited to build things with cloud code in a 0:14:02.967 --> 0:14:05.967 way that's not really hitting for me. Uh, on the 0:14:05.967 --> 0:14:12.287 other platforms, my secondary backup for cloud code is still cursor. Um, 0:14:12.286 --> 0:14:15.687 I was messing with, uh, windsurf for a while and 0:14:15.687 --> 0:14:18.167 some other tools, and I still dabble with all the 0:14:18.166 --> 0:14:21.527 other tools when I see something, uh, come out or new, 0:14:21.527 --> 0:14:24.927 I go and tinker with it. But right now, cloud 0:14:24.927 --> 0:14:29.727 code is my main, with cursor being secondary. I'm also 0:14:29.727 --> 0:14:33.207 switched over to using Dia as my main daily driver 0:14:33.207 --> 0:14:37.567 for my browser. And um, I'm on all the Mac 0:14:37.567 --> 0:14:43.206 OS 26 betas. They are fantastic. And uh, specifically having 0:14:43.207 --> 0:14:47.847 your phone app on the desktop is fantastic. And, uh, 0:14:47.847 --> 0:14:51.807 the podcast app is much improved. You can actually go 0:14:51.847 --> 0:14:53.527 above the two x limit. You can go up to 0:14:53.527 --> 0:14:57.217 three x now. And yeah, overall I would say Apple 0:14:57.217 --> 0:15:01.976 just really did a lot of quality updates. For example 0:15:02.017 --> 0:15:07.137 AirPod handoffs are much better. So that's it for updates 0:15:07.137 --> 0:15:11.017 I think. And let's get into cybersecurity. Apple quietly fixed 0:15:11.017 --> 0:15:16.377 an iPhone zero Day that was used against journalists. Echo 0:15:16.377 --> 0:15:22.857 leak using uses markdown syntax to bypass Microsoft 365 Copilot security. 0:15:23.337 --> 0:15:25.897 So some researchers found a smart way to steal data 0:15:25.897 --> 0:15:31.456 from Copilot by using obscure markdown link formats that Microsoft 0:15:31.457 --> 0:15:37.697 did not filter. Researchers turn 2 a.m. Tokyo hotel room 0:15:37.697 --> 0:15:43.537 chat into a Netflix RC. So Chubbs, who is a 0:15:43.897 --> 0:15:51.337 friend of mine and very, very good, um, attacker, uh, attacker. Um, 0:15:51.817 --> 0:15:56.237 what are we saying? Researcher. Bug. Bounty guy. Um. Security 0:15:56.277 --> 0:16:00.677 guy focused on offensive security and automated testing. He's also 0:16:00.717 --> 0:16:04.317 the creator of Asset note. So he's just one of 0:16:04.317 --> 0:16:08.157 the top, you know, top 1% or 1% of the 1%. 0:16:08.477 --> 0:16:10.837 I would put him in Sam Curry like up in 0:16:10.877 --> 0:16:15.917 that top, top echelon. And, uh, basically they found, uh, 0:16:17.797 --> 0:16:21.237 the way they found a way to attack an unclaimed 0:16:21.237 --> 0:16:25.677 internal package name called NFC logger. And they use that 0:16:25.677 --> 0:16:30.517 to get Archie actually on a Netflix, uh, system or 0:16:30.517 --> 0:16:34.757 within Netflix itself. And I'm sure they fixed that. Otherwise 0:16:34.757 --> 0:16:39.037 we wouldn't be talking about it. All right. Going to 0:16:39.037 --> 0:16:43.477 skip over the Tliose sessions. Uh, offer. That was in 0:16:43.477 --> 0:16:46.797 the thing. Uh, don't really want to talk about that. Um, 0:16:47.157 --> 0:16:51.087 HTML spec now escapes angle brackets and attributes. Tributes, so 0:16:51.087 --> 0:16:53.887 Chrome and other browsers are now going to start escaping 0:16:54.847 --> 0:16:59.647 left and right. Caret characters in HTML attributes to prevent 0:16:59.647 --> 0:17:06.647 mutation XSS attacks Interpol dismantles 20,000 malicious IPS linked to 0:17:06.687 --> 0:17:13.167 69 malware variants. Cursor security rules project tackles unsafe AI 0:17:13.207 --> 0:17:20.206 generated code, Europol says Europol yeah, Europol says stolen data 0:17:20.207 --> 0:17:24.526 has become the new underground currency. Phishing messages this is 0:17:24.527 --> 0:17:27.927 a quote created by Llms have higher success rate than 0:17:27.927 --> 0:17:31.567 those written by humans. And this is a euro pool. 0:17:33.367 --> 0:17:41.806 Io CTA report US airlines quietly selling flight data to DHS. Yeah. 0:17:41.847 --> 0:17:45.527 Government agencies can search 39 months of flight data, including 0:17:45.527 --> 0:17:50.346 passenger names, itineraries, travel dates and credit card numbers, and 0:17:50.347 --> 0:17:54.786 other people buying this data are include like the Secret Service, SEC, DEA, 0:17:55.147 --> 0:18:01.427 and US Marshals Service. And that's in addition to the 0:18:01.466 --> 0:18:06.267 DHS and all its various services. Amazon's AI agents can 0:18:06.267 --> 0:18:12.187 build cyber defense signatures in minutes. National security thoughts on 0:18:12.547 --> 0:18:16.067 Israel's action against Iran. I'm actually not going to go 0:18:16.067 --> 0:18:22.107 super into this one. It's an evolving situation. As everyone knows, 0:18:22.427 --> 0:18:26.907 new stuff is happening like every few hours on this, uh, go, 0:18:26.907 --> 0:18:28.627 go read the newsletter if you want to see my 0:18:28.627 --> 0:18:33.867 take on this. Army is commissioning Big Tech executives as 0:18:33.907 --> 0:18:38.547 lieutenant colonel. So for Silicon Valley executives from meta, Palantir 0:18:38.907 --> 0:18:43.266 and OpenAI are being brought in as lieutenant colonels to 0:18:43.307 --> 0:18:47.997 speed up the government's adoption of technology. And I'm overall 0:18:47.997 --> 0:18:52.677 for this, but I also have my spider senses going 0:18:52.677 --> 0:18:58.596 off in terms of like dystopia type situation. Again, I, 0:18:59.357 --> 0:19:03.956 I see the value of Palantir. I see the value 0:19:03.956 --> 0:19:10.077 of winning at almost any cost against China. Right? That 0:19:10.077 --> 0:19:12.436 is why I support this type of move from the 0:19:12.436 --> 0:19:17.877 military and why I support overall. In some ways, companies 0:19:17.877 --> 0:19:23.917 like Palantir and Andriel and you know why these companies exist? Yes, 0:19:23.917 --> 0:19:29.077 we need to move super fast. Unfortunately, the cultures and 0:19:29.077 --> 0:19:34.957 the ambitions and a whole bunch of externalities often come 0:19:35.277 --> 0:19:40.077 with moving really fast and in really ambitious ways. In 0:19:40.117 --> 0:19:43.797 with technology companies where you're making tons of money supporting 0:19:43.797 --> 0:19:47.456 the military. I don't think I need to spell it out. 0:19:47.696 --> 0:19:52.976 There's possibility for harm here, right? There's possibility for harm here. 0:19:53.177 --> 0:19:57.537 And I think we have to take that risk. I 0:19:57.537 --> 0:19:59.897 think this is the right move to take that risk. 0:20:00.177 --> 0:20:03.137 But holy crap, is it scary? Yes. And do we 0:20:03.137 --> 0:20:06.937 need oversight? Yes. Do we need to be extremely careful? Yes. 0:20:06.976 --> 0:20:09.417 Do we need to not turn our eyes away and 0:20:09.417 --> 0:20:15.737 just hope it all works out? Absolutely. Cheap drones will 0:20:15.736 --> 0:20:21.457 massively disrupt current large military dominance. Yeah. Israel and Ukraine 0:20:21.456 --> 0:20:26.057 are showing that $500 drones can destroy extremely expensive military gear, 0:20:26.456 --> 0:20:31.137 which could upend the advantage of countries like, you know, Russia, 0:20:31.137 --> 0:20:39.016 the US. We're already seeing it with Russia in Ukraine. Uh, 0:20:39.017 --> 0:20:42.016 big scope aside on this one. I'm curious about how 0:20:42.017 --> 0:20:45.506 this is going to start trickling into consumer and personal security. 0:20:45.507 --> 0:20:48.827 Like when will executive production teams all need to have 0:20:48.867 --> 0:20:53.226 like Anti-drone tech as part of their package? I think 0:20:53.267 --> 0:20:58.347 probably sooner rather than later. I that apple paper saying 0:20:58.387 --> 0:21:02.667 eyes don't reason was highly flawed, and someone named Alex 0:21:02.706 --> 0:21:06.706 Lawson says that the paper basically didn't do a good 0:21:06.706 --> 0:21:11.226 job and it was easy to make better examples. Although 0:21:11.226 --> 0:21:15.507 I've heard someone did a paper like this, this might 0:21:15.547 --> 0:21:18.106 have been Alex Lawson. Actually, I don't know if it 0:21:18.107 --> 0:21:22.826 was this paper. I heard somebody say that that whole paper, 0:21:22.827 --> 0:21:27.466 this whole takedown paper was actually Claude generated, and it 0:21:27.466 --> 0:21:30.907 wasn't a real paper. And it was just like, I 0:21:30.907 --> 0:21:33.787 don't know, fake or hype or whatever, and that they 0:21:33.787 --> 0:21:36.987 didn't expect it to blow up so much. I don't 0:21:36.986 --> 0:21:40.186 know if it was this one, but interesting to note. 0:21:40.547 --> 0:21:42.706 Also interesting to note, and I'm going to beat up 0:21:42.706 --> 0:21:47.686 on myself here for a second. I don't know if 0:21:47.686 --> 0:21:51.007 this was the paper that said that it doesn't actually 0:21:51.007 --> 0:21:55.047 matter if I know, or it doesn't actually matter if 0:21:55.047 --> 0:21:58.167 it was the paper or not. The fact that I 0:21:58.167 --> 0:22:03.847 put it in the newsletter is potentially a strike against me. 0:22:04.087 --> 0:22:09.407 Think about this. I counted as a strike against myself. Um, 0:22:09.726 --> 0:22:12.607 the thing is, I'm not sure if I would do 0:22:12.607 --> 0:22:16.686 it differently. It's just maybe I, I don't know. This 0:22:16.686 --> 0:22:19.167 is weird. Okay. Because when I put something in the 0:22:19.167 --> 0:22:21.927 newsletter and I talk about it, how much of a 0:22:21.927 --> 0:22:26.686 high how much have I endorsed this thing and said 0:22:26.686 --> 0:22:29.246 for sure that this is a real paper? For sure. 0:22:29.247 --> 0:22:33.407 These are real results because I read the paper fully. 0:22:33.807 --> 0:22:35.806 Not only that, but I looked at the all the 0:22:35.807 --> 0:22:41.407 testing methodologies and I evaluated the data and I confirmed 0:22:41.696 --> 0:22:44.897 that their testing was not a lie. That is a 0:22:44.936 --> 0:22:47.857 level of depth that I cannot do. For the amount 0:22:47.857 --> 0:22:51.777 of papers and stories that I'm doing. So what I'm 0:22:51.777 --> 0:22:55.737 saying is it is possible for me to be duped. Um, 0:22:55.777 --> 0:22:59.137 and this one very well could be, uh, in fact, 0:22:59.177 --> 0:23:00.417 you know what? I'm going to click on a link 0:23:00.417 --> 0:23:07.097 here and see if there's possible, like updates. I don't 0:23:07.097 --> 0:23:09.976 think so. I don't think so. I don't think it applied. 0:23:09.976 --> 0:23:12.857 It doesn't actually matter if it applied. This paper that 0:23:12.857 --> 0:23:16.496 I included might be completely legit. And I think it is, 0:23:16.497 --> 0:23:19.537 but it doesn't actually matter because this strike against me 0:23:19.577 --> 0:23:23.537 still counts, right? So this is a question from me 0:23:23.537 --> 0:23:27.857 to myself and also from you to me and from 0:23:27.857 --> 0:23:32.777 me to you, to me. You got to watch what 0:23:32.777 --> 0:23:35.417 people are posting, and you got to watch with how 0:23:35.456 --> 0:23:38.977 sure that they are that that thing is actually real 0:23:40.117 --> 0:23:42.877 And whether or not they're just using something that sounds 0:23:42.877 --> 0:23:47.316 real to support their own opinion. This one happened to 0:23:47.357 --> 0:23:50.437 go with my opinion. Oh, so what do you know? Magically, 0:23:50.517 --> 0:23:55.037 I just included in the newsletter as support for my opinion. Well, 0:23:55.037 --> 0:23:57.956 what if it turns out there was complete garbage? Well 0:23:57.956 --> 0:24:01.476 then I'm just like anyone else who is posting a 0:24:01.476 --> 0:24:05.236 random garbage. I would argue I'm not like that, but 0:24:05.277 --> 0:24:07.757 I'm saying it looks kind of the same, so I 0:24:07.757 --> 0:24:11.277 don't really care how much the difference is. So just 0:24:11.277 --> 0:24:14.037 keep in mind. And not just for me, but for yourself. 0:24:14.677 --> 0:24:17.877 You got to watch for whether or not evidence that's 0:24:17.877 --> 0:24:22.357 coming in is only air quote evidence, because how much 0:24:22.476 --> 0:24:26.117 time have you actually spent looking at the evidence, redoing 0:24:26.117 --> 0:24:31.597 the paper yourself, you know. And so how much are 0:24:31.597 --> 0:24:37.637 we using external validation or citations pointing to a paper 0:24:37.637 --> 0:24:43.216 Or as support for the paper being legit. And I 0:24:43.257 --> 0:24:46.577 think it gets more and more serious the further we go. Um, 0:24:46.617 --> 0:24:49.657 the good news is, a big part of my personal 0:24:49.936 --> 0:24:53.897 AI structure is going to be to find stuff like this. Uh, 0:24:53.897 --> 0:24:55.737 I want to be able to find stuff like this 0:24:55.736 --> 0:24:59.177 with my AI infrastructure where it's like, I don't know, 0:24:59.417 --> 0:25:01.777 seems like they use a lot of, uh, catchy AI 0:25:01.817 --> 0:25:04.137 terms in here, and it looks like they were really 0:25:04.177 --> 0:25:07.457 hand-wavy with their examples. Like, I want to be able 0:25:07.456 --> 0:25:10.617 to catch this stuff for myself. Just something to keep 0:25:10.617 --> 0:25:16.657 in mind. OpenAI's O3 Pro is very smart, but context hungry. Agreed. 0:25:17.177 --> 0:25:21.537 Man killed by police after spiraling into a ChatGPT driven psychosis. 0:25:22.657 --> 0:25:26.737 Google plans to cut ties with scale AI because basically 0:25:26.736 --> 0:25:30.337 meta bought like half of them or something. CIO wants 0:25:30.337 --> 0:25:34.897 to clone staff as digital twins and AI agents. UC 0:25:34.897 --> 0:25:39.067 San Diego CIO wants to extract knowledge from experienced IT 0:25:39.107 --> 0:25:45.907 staff and create digital twins that handle routine problems. Well, yeah, obviously, 0:25:46.027 --> 0:25:50.667 you know, you got somebody, uh, greybeard, you know, Chris Myers, 0:25:51.147 --> 0:25:54.667 who's been at the company for 41 years, and they 0:25:54.667 --> 0:25:58.267 know everything. And anyone who, you know, finds something broken 0:25:58.267 --> 0:26:01.867 that nobody could fix, they all call Chris Myers. So 0:26:01.867 --> 0:26:05.586 the CEO is like, let's go find Chris Myers. Let's 0:26:05.587 --> 0:26:10.067 go interview him for seven weeks in a dark, padded 0:26:10.067 --> 0:26:12.987 room and pull out all of his knowledge and put 0:26:12.986 --> 0:26:15.667 that into an AI agent. And now you could just 0:26:15.706 --> 0:26:18.947 ask a virtual Chris Myers. Well, obviously we need this 0:26:18.946 --> 0:26:21.347 for everything, right? We need this for the best doctors. 0:26:21.347 --> 0:26:24.387 We need this for the best psychiatrists. We need this 0:26:24.387 --> 0:26:27.627 for everything. This is you know, this is the problem. When, 0:26:28.107 --> 0:26:31.987 you know, extremely experienced people leave a company, you lose 0:26:31.986 --> 0:26:36.087 all the knowledge that they, uh, took with them. And 0:26:36.087 --> 0:26:41.966 unfortunately also for I, it's also knowledge that you can't 0:26:41.966 --> 0:26:47.647 possibly extract from them efficiently, right? You can't possibly ask 0:26:47.647 --> 0:26:51.726 enough questions to to gather their knowledge. Their knowledge is, 0:26:52.486 --> 0:26:56.887 let's call it infinite. 41 years of knowledge essentially baked 0:26:56.887 --> 0:27:01.726 into an AI model. That's a great analogy actually it. 0:27:02.206 --> 0:27:06.167 How many questions do you have to ask? You know, uh, 0:27:06.647 --> 0:27:11.367 O3 to get all of O3 knowledge out of it? 0:27:11.686 --> 0:27:15.046 The answer is an infinite number of questions. So it 0:27:15.047 --> 0:27:18.367 doesn't matter how much you interview Chris Myers, who's been 0:27:18.367 --> 0:27:21.686 there for 41 years, you're not going to get enough. Right? 0:27:21.726 --> 0:27:28.327 So so this is a serious challenge. ChatGPT dominates LLM 0:27:28.367 --> 0:27:31.927 usage at 80%, 86% market share. I did not know 0:27:31.927 --> 0:27:35.216 it was that high technology fabric. Summary of the massive 0:27:35.257 --> 0:27:39.297 Google outage. Yeah. This is this is really cool. Um, 0:27:40.337 --> 0:27:42.737 so they put out the, you know, Google put out 0:27:42.736 --> 0:27:47.617 their their giant paper saying what happened. And when I see, 0:27:47.657 --> 0:27:50.817 you know, a 14 page paper or whatever. And it's 0:27:50.817 --> 0:27:53.817 about an outage and describing an outage, it's not like 0:27:53.817 --> 0:27:57.377 deep human knowledge that I'm getting. That's where I run Extraction's. 0:27:57.736 --> 0:28:03.216 So I've got this, um, pattern called create story explanation, 0:28:03.696 --> 0:28:09.377 create underscore story story underscore explanation. It is one of 0:28:09.377 --> 0:28:14.337 my favorite patterns. It explains anything at like a ninth 0:28:14.337 --> 0:28:17.897 grade level in kind of a story format. So this happened, 0:28:17.897 --> 0:28:22.296 then this happened, then this happened. And it's just very clear. 0:28:22.456 --> 0:28:26.097 So I recommend checking it out if you care about 0:28:26.097 --> 0:28:28.977 that outage, but more importantly using it against other stuff. 0:28:29.537 --> 0:28:33.797 YouTube officially beats all other streaming platforms. In US viewership, 0:28:34.157 --> 0:28:40.437 they have 13% of all US TV viewership. Oh, wow. Okay. 0:28:40.477 --> 0:28:44.277 I thought it was 13% of streaming. Okay, I read 0:28:44.277 --> 0:28:52.117 that wrong. 13% of all US TV viewership. That's insane. 0:28:55.157 --> 0:29:00.877 That's insane. Okay, YouTube by itself, 13% of TV. Yeah, 0:29:00.917 --> 0:29:03.037 it seems very high. I bet you that goes up 0:29:03.077 --> 0:29:07.437 crazy amounts in the next few years. It's basically the 0:29:07.437 --> 0:29:09.837 only TV that I watched. I was actually at a 0:29:09.837 --> 0:29:14.437 resort yesterday in Napa, and they only had regular TV, 0:29:14.437 --> 0:29:18.036 and it was like it was like a commercial rectangle. 0:29:19.197 --> 0:29:22.197 All I did was watch commercials and occasionally they would 0:29:22.197 --> 0:29:26.237 show like part of a show. Anyway, Amazon joins the 0:29:26.237 --> 0:29:34.207 big nuclear party buying 1.2GW, uh, for AWS. tvOS 26 0:29:34.527 --> 0:29:37.847 hints at built in camera for a new Apple TV 4K. 0:29:38.567 --> 0:29:40.727 I wish it was an Apple TV 8-K, and I 0:29:40.727 --> 0:29:43.967 wish it had ten Gigabit Ethernet. That's what I wish. 0:29:44.127 --> 0:29:45.887 But I would like to see a camera because I'd 0:29:45.887 --> 0:29:49.727 like to do FaceTime with, um, in the living room 0:29:49.727 --> 0:29:52.447 sitting on the couch. That would be nice. I also 0:29:52.447 --> 0:29:56.527 don't want a camera facing the living room, so I 0:29:56.527 --> 0:30:01.447 would need one of those flip dongle cover things. Apple's 0:30:01.447 --> 0:30:05.207 iPad OS 26 finally makes the iPad a real computer. 0:30:05.207 --> 0:30:09.967 Everyone is talking about this. People are saying that this 0:30:10.007 --> 0:30:14.887 actually allows you to use an iPad as a laptop, 0:30:15.407 --> 0:30:19.247 almost like very closely. And it's kind of been universal 0:30:19.247 --> 0:30:22.087 that people are saying, Holy crap, this is really, really good. 0:30:22.687 --> 0:30:26.007 I have not. I don't use my iPads much. And 0:30:26.087 --> 0:30:29.947 the reason is I just didn't like the OS. The 0:30:29.947 --> 0:30:32.347 only thing that I use it for is reading and 0:30:32.347 --> 0:30:36.187 drawing and sketching and doing stuff like that with like freeform. 0:30:36.667 --> 0:30:39.187 And for that I really love it. And for reading, 0:30:39.187 --> 0:30:41.227 I really love it. But I don't use it as 0:30:41.227 --> 0:30:44.787 a computer because I prefer laptops. Maybe that will change 0:30:46.107 --> 0:30:48.667 the argument that it's time to kill Siri. I did 0:30:48.667 --> 0:30:52.147 not know that I believe this, but I definitely do. 0:30:52.187 --> 0:30:57.267 I think once Apple fixes its AI interface, its universal 0:30:57.267 --> 0:31:00.747 AI to life OS that I've been talking about, they 0:31:00.747 --> 0:31:05.947 need to call it something new. Siri has too much baggage. I. 0:31:07.547 --> 0:31:08.787 I can show them if you ask. 0:31:09.307 --> 0:31:15.067 Hey Siri, stop the one time Siri actually listens when 0:31:15.067 --> 0:31:20.867 I'm trying to record, um, it essentially that system does 0:31:20.867 --> 0:31:24.907 not turn on and turn off lights reliably. It has 0:31:24.907 --> 0:31:28.277 so many problems. It's been over ten years. It had 0:31:28.277 --> 0:31:31.597 its attempt. It had its chances, and now it needs 0:31:31.597 --> 0:31:34.757 a new name. I thoroughly believe this. It's just a 0:31:34.757 --> 0:31:38.397 marketing concept. When you have enough bad baggage associated with something, 0:31:38.797 --> 0:31:41.237 you got to flip it over. You got to retire it. 0:31:41.957 --> 0:31:45.997 Nvidia writes off China revenue in company forecasts because, you 0:31:45.997 --> 0:31:50.157 know it's too unpredictable. With the current administration, Waymo rides 0:31:50.157 --> 0:31:52.997 cost more than Uber or Lyft, but people are happy 0:31:52.997 --> 0:31:56.677 paying it so often 5 or $10 more for a 0:31:56.717 --> 0:32:00.757 ride versus like Uber or a taxi. But people are 0:32:00.757 --> 0:32:03.277 still paying it and they're still making money because it's 0:32:03.277 --> 0:32:08.036 so much of a better experience. Humans. Trump ends protection 0:32:08.037 --> 0:32:12.157 for Afghans as Congress scrambles to intervene. I'm really upset 0:32:12.157 --> 0:32:18.397 about this, really upset that these Afghanis who had helped 0:32:18.397 --> 0:32:22.597 the US military during the war and who cannot go 0:32:22.597 --> 0:32:26.537 back because they are being hunted by the Taliban and 0:32:26.537 --> 0:32:32.857 because their families are under like watch by the Taliban, 0:32:32.857 --> 0:32:36.336 they are in grave danger. Their families are and they are. 0:32:36.537 --> 0:32:38.296 And if they go back and try to link with 0:32:38.297 --> 0:32:41.897 their families because they're forcibly kicked out of the US, 0:32:42.897 --> 0:32:47.977 it is essentially damning them and their families, potentially to die. 0:32:48.857 --> 0:32:53.297 And this is after they served and sacrificed essentially for 0:32:53.297 --> 0:32:56.817 the US military. I do not believe we should treat 0:32:56.817 --> 0:33:03.217 allies like that. Robin I system makes first autonomous scientific discovery. 0:33:03.217 --> 0:33:06.857 So this is future houses. Robin I so I talked 0:33:06.857 --> 0:33:13.177 about this story of automated, you know, science assistance. And 0:33:13.177 --> 0:33:16.017 this is even more it's more end to end the 0:33:16.017 --> 0:33:21.737 entire research process. So it, um, it discovered that ripasudil 0:33:22.057 --> 0:33:28.306 could treat dry macular degeneration by orchestrating multiple specialized agents 0:33:28.307 --> 0:33:33.347 to handle the entire research process autonomously in just 2.5 months. 0:33:34.107 --> 0:33:37.667 And who knows how much of that is like the 0:33:37.667 --> 0:33:41.747 entire air quotes entire research process, but supposedly it went 0:33:41.747 --> 0:33:46.546 end to end and actually worked. So this is the 0:33:46.547 --> 0:33:49.347 type of thing that I think is really, really exciting. 0:33:50.067 --> 0:33:55.067 Ultra black paint might solve satellite light pollution crisis. Although 0:33:55.067 --> 0:33:57.907 I'm curious, wouldn't you just see black dots moving through 0:33:57.947 --> 0:34:03.347 your pictures, blocking out stars and, you know, the moon 0:34:03.347 --> 0:34:07.987 or whatever? I guess the black dots wouldn't be as annoying. 0:34:09.867 --> 0:34:12.667 It seems like they could still overwrite stars. If you 0:34:12.707 --> 0:34:14.907 are trying to take a picture of the starry sky, 0:34:15.627 --> 0:34:18.787 you might still see artifacts, but it won't be nearly 0:34:18.787 --> 0:34:22.967 as bad as like a bright white line Him drawing 0:34:22.967 --> 0:34:28.447 over your picture, actually multiple bright white lines. But, uh, 0:34:28.447 --> 0:34:31.727 I wonder how possible this is, like, are we going 0:34:31.727 --> 0:34:33.727 to pull all the ones that are up there down? 0:34:33.727 --> 0:34:37.647 Are we going to paint the ones that are up there? Doubtful. 0:34:38.967 --> 0:34:41.287 So this would involve painting all the new ones that 0:34:41.287 --> 0:34:46.367 are going up, which unfortunately is many thousands. And the 0:34:46.367 --> 0:34:49.327 Pentagon has evidently been pushing Americans to believe in UFOs 0:34:49.327 --> 0:34:53.887 for decades. And evidently, according to this report, this is 0:34:53.887 --> 0:34:57.487 to cover up actual secret weapons programs. So they basically 0:34:57.487 --> 0:35:00.767 circulate this stuff on purpose to get people talking about 0:35:01.327 --> 0:35:07.967 a red herring discovery. Fiddle ITM brings malicious traffic detection 0:35:07.967 --> 0:35:10.327 to man of the middle proxy. Oh, I'm not going 0:35:10.327 --> 0:35:11.847 to actually read all of these, so I'm just going 0:35:11.887 --> 0:35:14.047 to scroll through and look for some really cool ones, 0:35:14.167 --> 0:35:17.007 or if any are from my friends. Let's see. Boom 0:35:17.007 --> 0:35:20.407 boom boom boom. Oh, Sam Altman's gentle singularity. You probably 0:35:20.407 --> 0:35:22.417 want to read this one. I've also got a fabric 0:35:22.457 --> 0:35:25.297 extracted predictions that comes out of it in the newsletter, 0:35:25.297 --> 0:35:34.137 so check that out. And yeah, I think that's about it. Okay, 0:35:34.137 --> 0:35:36.057 this is the end of the standard edition of the podcast, 0:35:36.057 --> 0:35:38.257 which includes just the news items for the week to 0:35:38.297 --> 0:35:40.377 get the rest of the episode, which includes much more 0:35:40.377 --> 0:35:43.817 of my analysis, the ideas section, and the weekly member essay. 0:35:43.817 --> 0:35:46.497 Please consider becoming a member. As a member, you get 0:35:46.537 --> 0:35:49.297 access to all sorts of stuff, most importantly, access to 0:35:49.297 --> 0:35:52.457 our extraordinary community of over a thousand brilliant and kind 0:35:52.497 --> 0:35:56.897 people in industries like cybersecurity, AI, and the humanities. You 0:35:56.897 --> 0:35:59.577 also get access to the UL Book Club, dedicated member 0:35:59.577 --> 0:36:02.577 content and events, and lots more. Plus, you'll get a 0:36:02.577 --> 0:36:04.897 dedicated podcast feed that you can put into your client 0:36:04.897 --> 0:36:07.417 that gets you the full member edition of the podcast 0:36:07.417 --> 0:36:09.457 that basically doesn't have this in it, and just goes 0:36:09.457 --> 0:36:11.937 all the way through with all the different sections. So 0:36:11.977 --> 0:36:13.817 to become a member and get all that, just head 0:36:13.817 --> 0:36:18.737 over to Daniel Music.com. That's Daniel Miessler, and we'll see 0:36:18.737 --> 0:36:19.377 you next time.